Nunnari LabsResearchSeptember 5, 2026
Fitting three vision models into two megabytes
Design notes from an on-device smart-glasses programme on the Qualcomm QCS6490 — why three concurrent models on one Hexagon NPU is slower than one at a time, and what an operator-constrained NPU does to a vision-language architecture.
Nunnari LabsResearchSeptember 5, 2026
Two OCR engines and a learning loop that needs no code change
Engineering notes from ARB Corporation's invoice pipeline on Oracle Cloud — why one OCR engine was not enough, how per-supplier sub-prompts close the human-in-the-loop loop, and how a GPU node pool that scales to zero keeps the cost honest.
Nunnari LabsResearchAugust 15, 2026
Validating NVIDIA AI Enterprise on L40S before production
The three-day run sheet we use to bring up L40S GPU servers under NVIDIA AI Enterprise — what has to be proven on day one before any model is benchmarked, and how the benchmark days are structured.
Nunnari LabsNewsJanuary 20, 2025
Nunnari Labs achieves ISO/IEC 42001:2023 certification
Nunnari Labs' AI management system is certified to ISO/IEC 42001:2023, the first international standard for governing AI across its lifecycle. What the audit covered and what it means for the work we ship.
Nunnari LabsPerspectiveMarch 25, 2023
Harnessing the value of generative AI
Generative AI is rapidly transforming industries, unlocking new levels of operational efficiency, customer engagement and innovation.